# Установка библиотек

!pip install phik
!pip install --upgrade scikit-learn
!pip install catboost

Requirement already satisfied: phik in /Users/NOrlov/myenv/lib/python3.12/site-packages (0.12.4)
Requirement already satisfied: numpy>=1.18.0 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from phik) (1.26.4)
Requirement already satisfied: scipy>=1.5.2 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from phik) (1.13.0)
Requirement already satisfied: pandas>=0.25.1 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from phik) (2.2.1)
Requirement already satisfied: matplotlib>=2.2.3 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from phik) (3.8.4)
Requirement already satisfied: joblib>=0.14.1 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from phik) (1.3.2)
Requirement already satisfied: contourpy>=1.0.1 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from matplotlib>=2.2.3->phik) (1.2.1)
Requirement already satisfied: cycler>=0.10 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from matplotlib>=2.2.3->phik) (0.12.1)
Requirement already satisfied: fonttools>=4.22.0 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from matplotlib>=2.2.3->phik) (4.50.0)
Requirement already satisfied: kiwisolver>=1.3.1 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from matplotlib>=2.2.3->phik) (1.4.5)
Requirement already satisfied: packaging>=20.0 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from matplotlib>=2.2.3->phik) (24.0)
Requirement already satisfied: pillow>=8 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from matplotlib>=2.2.3->phik) (10.3.0)
Requirement already satisfied: pyparsing>=2.3.1 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from matplotlib>=2.2.3->phik) (3.1.2)
Requirement already satisfied: python-dateutil>=2.7 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from matplotlib>=2.2.3->phik) (2.9.0.post0)
Requirement already satisfied: pytz>=2020.1 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from pandas>=0.25.1->phik) (2024.1)
Requirement already satisfied: tzdata>=2022.7 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from pandas>=0.25.1->phik) (2024.1)
Requirement already satisfied: six>=1.5 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from python-dateutil>=2.7->matplotlib>=2.2.3->phik) (1.16.0)
Requirement already satisfied: scikit-learn in /Users/NOrlov/myenv/lib/python3.12/site-packages (1.4.2)
Requirement already satisfied: numpy>=1.19.5 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from scikit-learn) (1.26.4)
Requirement already satisfied: scipy>=1.6.0 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from scikit-learn) (1.13.0)
Requirement already satisfied: joblib>=1.2.0 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from scikit-learn) (1.3.2)
Requirement already satisfied: threadpoolctl>=2.0.0 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from scikit-learn) (3.4.0)
Requirement already satisfied: catboost in /Users/NOrlov/myenv/lib/python3.12/site-packages (1.2.3)
Requirement already satisfied: graphviz in /Users/NOrlov/myenv/lib/python3.12/site-packages (from catboost) (0.20.3)
Requirement already satisfied: matplotlib in /Users/NOrlov/myenv/lib/python3.12/site-packages (from catboost) (3.8.4)
Requirement already satisfied: numpy>=1.16.0 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from catboost) (1.26.4)
Requirement already satisfied: pandas>=0.24 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from catboost) (2.2.1)
Requirement already satisfied: scipy in /Users/NOrlov/myenv/lib/python3.12/site-packages (from catboost) (1.13.0)
Requirement already satisfied: plotly in /Users/NOrlov/myenv/lib/python3.12/site-packages (from catboost) (5.20.0)
Requirement already satisfied: six in /Users/NOrlov/myenv/lib/python3.12/site-packages (from catboost) (1.16.0)
Requirement already satisfied: python-dateutil>=2.8.2 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from pandas>=0.24->catboost) (2.9.0.post0)
Requirement already satisfied: pytz>=2020.1 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from pandas>=0.24->catboost) (2024.1)
Requirement already satisfied: tzdata>=2022.7 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from pandas>=0.24->catboost) (2024.1)
Requirement already satisfied: contourpy>=1.0.1 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from matplotlib->catboost) (1.2.1)
Requirement already satisfied: cycler>=0.10 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from matplotlib->catboost) (0.12.1)
Requirement already satisfied: fonttools>=4.22.0 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from matplotlib->catboost) (4.50.0)
Requirement already satisfied: kiwisolver>=1.3.1 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from matplotlib->catboost) (1.4.5)
Requirement already satisfied: packaging>=20.0 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from matplotlib->catboost) (24.0)
Requirement already satisfied: pillow>=8 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from matplotlib->catboost) (10.3.0)
Requirement already satisfied: pyparsing>=2.3.1 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from matplotlib->catboost) (3.1.2)
Requirement already satisfied: tenacity>=6.2.0 in /Users/NOrlov/myenv/lib/python3.12/site-packages (from plotly->catboost) (8.2.3)

# Импорт библиотек

import pandas as pd

import numpy as np

import matplotlib.pyplot as plt

import seaborn as sns

import plotly.graph_objects as go
import plotly.express as px

from sklearn.metrics import root_mean_squared_error, mean_squared_error
from sklearn.preprocessing import StandardScaler, OneHotEncoder, LabelEncoder
from sklearn.model_selection import train_test_split, GridSearchCV, cross_val_score, TimeSeriesSplit
from sklearn.linear_model import LinearRegression
from sklearn.dummy import DummyRegressor
from sklearn.compose import ColumnTransformer
from sklearn.pipeline import Pipeline

from phik import phik_matrix

from catboost import CatBoostRegressor

import lightgbm as lgb

from statsmodels.tsa.seasonal import seasonal_decompose
from statsmodels.graphics.tsaplots import plot_acf

import time

import os

# Загрузка данных

try:
    path = '/datasets/taxi.csv'
    data = pd.read_csv(path, index_col=[0], parse_dates=[0])

except FileNotFoundError:
    current_directory = os.getcwd()
    data = pd.read_csv(current_directory + path, index_col=[0], parse_dates=[0])

display(data.info())

# Назначение индексов

data.sort_index(inplace=True)

# Поиск пропусков

print()
print('Пропуски:', data.isna().sum())
print()

#Ресэмплинг почасово

data = data.resample('1H').sum()

<class 'pandas.core.frame.DataFrame'>
DatetimeIndex: 26496 entries, 2018-03-01 00:00:00 to 2018-08-31 23:50:00
Data columns (total 1 columns):
 #   Column      Non-Null Count  Dtype
---  ------      --------------  -----
 0   num_orders  26496 non-null  int64
dtypes: int64(1)
memory usage: 414.0 KB

None

Пропуски: num_orders    0
dtype: int64

/var/folders/2r/zfmtb66x3rs14ctpbj7lhdt40000gp/T/ipykernel_19445/1415298997.py:15: FutureWarning: 'H' is deprecated and will be removed in a future version, please use 'h' instead.
  data = data.resample('1H').sum()

# Вывод датафрейма

display(data.sample(5))

# Визуализация

plt.figure(figsize=(16, 8))
plt.plot(data.index, data['num_orders'], linestyle='-', color='b')
plt.ylabel('Order number per hour')
plt.title('Order numbers per hour');

# Общие сведения о заказах

data.describe()

# Гистограмма

plt.figure(figsize = (16,8))
plt.hist(data, bins = 100)
plt.ylabel('Frequency')
plt.xlabel('Order numbers')
plt.title('Order number hist')
plt.grid(True);

# Boxplot

plt.figure(figsize = (16,8))
plt.boxplot(data)
plt.ylabel('Order numbers per hour')
plt.title('Order number hist')
plt.grid(True);

data[data['num_orders']>250].sample(5)

decomposed = seasonal_decompose(data['num_orders'].dropna())

fig = go.Figure()

fig.add_trace(go.Scatter(x=data['num_orders'].dropna().index, y=data['num_orders'].dropna(),
                    mode='lines',
                    name='OriginalData'))

fig.add_trace(go.Scatter(x = decomposed.resid.index, y = decomposed.resid,
                    mode='lines',
                    name='Resid'))

fig.add_trace(go.Scatter(x = decomposed.seasonal.index, y = decomposed.seasonal,
                    mode='lines',
                    name='Season'))

fig.add_trace(go.Scatter(x = decomposed.trend.index, y = decomposed.trend,
                    mode='lines',
                    name='Trend'))

# Фурье преобразование сезонной компоненты

T = 1   # Интервал между отсчетами 1 час
L = len(decomposed.seasonal)   # Длина датасета
t = np.arange(0, L*T, T)  # Временная ось, часы

# Выполнение FFT

N = len(data)
Y = np.fft.fft(decomposed.seasonal)
Y = Y/np.max(Y)
Y = Y[range(N//2)]  # Только положительные частоты

# Создание временной оси

frequencies = 1/np.arange(0, (1/(2*T)), (1/(2*T))/len(Y))


fig = go.Figure()
fig.add_trace(go.Scatter(x=frequencies, y=np.abs(Y), mode='lines', name='FFT'))
fig.update_layout(title='Анализ временных интервалов "сезонной" компоненты данных',
                  xaxis_title='Периоды, часы',
                  yaxis_title='Относительная амплитуда')
fig.show();

/var/folders/2r/zfmtb66x3rs14ctpbj7lhdt40000gp/T/ipykernel_19445/122445324.py:16: RuntimeWarning:

divide by zero encountered in divide

# Построение коррелограммы сезонной компоненты

corr_data_seasonal = decomposed.seasonal.dropna()

fig, ax = plt.subplots(figsize=(16, 8))

# Построение коррелограммы с использованием fig и ax

plot_acf(corr_data_seasonal, lags=24*14, ax=ax)  # lags - количество лагов для отображения

plt.xlabel('Лаги, часы')
plt.ylabel('Автокорреляция')
plt.grid(True)
plt.title('Коррелограмма остатков')

Text(0.5, 1.0, 'Коррелограмма остатков')

# Построение коррелограммы остаточной компоненты

corr_data_resid = decomposed.resid.dropna()

fig, ax = plt.subplots(figsize=(16, 8))

# Построение коррелограммы с использованием fig и ax

plot_acf(corr_data_resid, lags=24*14, ax=ax)  # lags - количество лагов для отображения

plt.xlabel('Лаги, часы')
plt.ylabel('Автокорреляция')
plt.grid(True)
plt.title('Коррелограмма остаточной компоненты')
print('RMS остаточной компоненты:', np.std(corr_data_resid))

RMS остаточной компоненты: 24.694084222676633

# Создадим функцию, которое добавляет столбцы с часами, числом, днем недели, месяцем, а также прошлые данные и скользящее среднее

def make_features(data, max_lag, rolling_mean_size):
    data['hour'] = data.index.hour
    data['day'] = data.index.day
    data['dayofweek'] = data.index.dayofweek

    for lag in range(1, max_lag + 1):
        data['lag_{}'.format(lag)] = data['num_orders'].shift(lag)

    data['rolling_mean'] = data['num_orders'].shift().rolling(rolling_mean_size).mean()

# Создаем новые фичи в датасете с помощью функции

make_features(data, 8, 24)

data_final = data.dropna()
data_final.sample(15)

# Рассмотрим распределение данных, сгруппированных по отдельным признакам

# по часам

hour = data_final.groupby(by = 'hour').mean()['num_orders']
plt.figure(figsize = (16,8))
plt.bar(hour.index, hour)
plt.title('Hour distribution')
plt.xlabel('Hour')
plt.ylabel('Mean order value')
plt.show()

# по дням недели

week = data_final.groupby(by = 'dayofweek').mean()['num_orders']
plt.figure(figsize = (16,8))
plt.bar(week.index, week)
plt.title('Day of Week distribution')
plt.xlabel('Day of Week')
plt.ylabel('Mean order value')
plt.show()

# по числу

day = data_final.groupby(by = 'day').mean()['num_orders']
plt.figure(figsize = (16,8))
plt.bar(day.index, day)
plt.title('Day')
plt.xlabel('Day')
plt.ylabel('Mean order value')
plt.show()

# Выбираем интересующие нас числовые признаки

interval_cols = ['num_orders']

# Вычисляем матрицу корреляции с использованием phik

phik_corr_matrix = phik_matrix(data_final[['num_orders', 'dayofweek', 'hour', 'day']], interval_cols=interval_cols)

plt.figure(figsize=(18, 10))
sns.heatmap(phik_corr_matrix, annot=True, cmap='coolwarm', vmin=-1, vmax=1)
plt.title('PhiK Correlation Coefficients Heatmap')
plt.show();

# Data split

train, test = train_test_split(data_final, shuffle=False, test_size=0.1, random_state=42)
train = train.dropna()

y_train = train['num_orders']
X_train = train.drop('num_orders', axis = 1)

y_test = test['num_orders']
X_test = test.drop('num_orders', axis = 1)

# Создаем модель CatBoostRegressor

model_cb = CatBoostRegressor(random_state = 41, verbose = False)

# Задаем сетку параметров для GridSearchCV

param_grid = {
    'learning_rate': [0.01, 0.03, 0.1],
    'depth': [3, 6, 8],
    'l2_leaf_reg': [0.1, 1, 5],
    'iterations': [500]
}

# Инициализируем TimeSeriesSplit для кросс-валидации временных рядов

tscv = TimeSeriesSplit(n_splits=3)

grid_search_cb = GridSearchCV(estimator=model_cb, param_grid=param_grid, scoring='neg_mean_squared_error', cv=tscv, verbose=1, n_jobs=-1, error_score='raise')
grid_search_cb.fit(X_train, y_train)
best_params_cb = grid_search_cb.best_params_
best_model_cb = grid_search_cb.best_estimator_

Fitting 3 folds for each of 27 candidates, totalling 81 fits

# Результаты обучения и получение RMSE метрик на тестовом и тренировочном наборах

print("Best Hyperparameters CatBoostRegressor:", best_params_cb)
print("Best CatBoostRegressor model CV RMSE:", best_model_cb.best_score_)

y_pred_train_cb = best_model_cb.predict(X_train)
rmse_train_cb = mean_squared_error(y_train, y_pred_train_cb, squared = False)
print("Best CatBoostRegressor model Train RMSE:", rmse_train_cb)
print("Best CatBoostRegressor model Train mean residual:", np.mean(y_train - y_pred_train_cb))

Best Hyperparameters CatBoostRegressor: {'depth': 3, 'iterations': 500, 'l2_leaf_reg': 1, 'learning_rate': 0.1}
Best CatBoostRegressor model CV RMSE: {'learn': {'RMSE': 18.228244040730306}}
Best CatBoostRegressor model Train RMSE: 18.228244040730313
Best CatBoostRegressor model Train mean residual: 0.0023369308604390755

/Users/NOrlov/myenv/lib/python3.12/site-packages/sklearn/metrics/_regression.py:483: FutureWarning:

'squared' is deprecated in version 1.4 and will be removed in 1.6. To calculate the root mean squared error, use the function'root_mean_squared_error'.

# Замер времени обучения лучшей модели

start_time = time.time()

time_model_cb = best_model_cb
time_model_cb.fit(X_train, y_train)
training_time_cb = time.time() - start_time

# Замер времени предсказания лучшей модели

start_time = time.time()
y_pred_cb = time_model_cb.predict(X_train)
prediction_time_cb = time.time() - start_time

print('CatBoostRegressor model training time, s:', training_time_cb)
print('CatBoostRegressor model prediction time, s:', prediction_time_cb)

CatBoostRegressor model training time, s: 0.29205894470214844
CatBoostRegressor model prediction time, s: 0.0012660026550292969

# Визуализация y_true, y_pred, resid и их разницы для наглядности

fig = go.Figure()

fig.add_trace(go.Scatter(x=X_train.index, y=y_train,
                    mode='lines',
                    name='Original data'))

fig.add_trace(go.Scatter(x = X_train.index, y = y_pred_train_cb,
                    mode='lines',
                    name='CB prediction'))

fig.add_trace(go.Scatter(x = X_train.index, y = decomposed.resid.loc[X_train.index],
                    mode='lines',
                    name='Resid'))

fig.add_trace(go.Scatter(x = X_train.index, y = decomposed.resid.loc[X_train.index] + y_pred_train_cb.mean() - y_pred_train_cb,
                    mode='lines',
                    name='Resid-error'))

fig.update_layout(xaxis_tickformat='%Y-%m-%d', xaxis_title='Datetime')
fig.update_layout(width=1000, height=800)
fig.show()

# Построение гистограммы остатков на Train наборе

plt.figure(figsize = (16,8))
plt.hist(y_train - y_pred_train_cb, bins=100, label = 'model residual')
plt.hist(decomposed.resid.loc[X_train.index], bins=100, alpha = 0.7, label = 'residual component')
plt.xlabel('y_true - y_pred_train')
plt.ylabel('Frequency')
plt.axvline(x=0, color='r', linestyle='-', linewidth=1, label = 'zero')
plt.axvline(x=np.mean(y_train - y_pred_train_cb), color='b', linestyle='-', linewidth=2, alpha = 0.2, label = 'residual mean')
plt.legend()
plt.title('CatBoostRegressor residual hist')
plt.grid('true')
plt.show();

# Поиск наиболее и наименее значимых фичей

all_feature_names = X_train.columns

feature_importances = best_model_cb.feature_importances_

feature_importance_df = pd.DataFrame({
    'Feature': all_feature_names,
    'Importance': feature_importances
})

# Максимальные признаки

feature_importance_df_max = feature_importance_df.sort_values(
    by='Importance', ascending=False
)
plt.figure(figsize=(12, 8))
plt.barh(feature_importance_df_max['Feature'], feature_importance_df_max['Importance'], color='skyblue')
plt.xlabel('Importance')
plt.ylabel('Features')
plt.title('Maximal Feature Importance CatBoostRegressor')
plt.gca().invert_yaxis()
plt.show()

# Задаем сетку параметров для GridSearchCV

params_lgb = {
    'boosting_type': 'gbdt',
    'objective': 'regression',
    'metric': 'rmse',
    'max_depth': -1,
    'min_child_samples': 20,
    'subsample': 0.8,
    'colsample_bytree': 0.8,
    'reg_alpha': 0.0,
    'reg_lambda': 0.0,
    'verbose': 0,
    'force_col_wise': True,
    'n_estimators' : 500
}

param_grid_lgb = {
    'num_leaves': [10, 20, 30],
    'learning_rate': [0.01, 0.03, 0.1]
}

# Создаем модель LightGBM

model_lgb = lgb.LGBMRegressor(**params_lgb, random_state=42)


grid_search_lgb = GridSearchCV(estimator=model_lgb, param_grid=param_grid_lgb, scoring='neg_mean_squared_error', cv=tscv, verbose=1, n_jobs=-1, error_score='raise')
grid_search_lgb.fit(X_train, y_train)
best_params_lgb = grid_search_lgb.best_params_
best_model_lgb = grid_search_lgb.best_estimator_

Fitting 3 folds for each of 9 candidates, totalling 27 fits

# Результаты обучения и получение RMSE метрик на тренировочном наборе

print("Best Hyperparameters LightGBM:", best_params_lgb)
print("Best LightGBM model CV RMSE:", best_model_cb.best_score_)

y_pred_train_lgb = best_model_lgb.predict(X_train)
rmse_train_lgb = mean_squared_error(y_train, y_pred_train_lgb, squared = False)

print("Best LightGBM model Train RMSE:", rmse_train_lgb)
print("Best LightGBM Train mean residual:", np.mean(y_train - y_pred_train_lgb))

Best Hyperparameters LightGBM: {'learning_rate': 0.01, 'num_leaves': 30}
Best LightGBM model CV RMSE: {'learn': {'RMSE': 18.228244040730306}}
Best LightGBM model Train RMSE: 17.493857826582488
Best LightGBM Train mean residual: 1.5362727913915024e-10

/Users/NOrlov/myenv/lib/python3.12/site-packages/sklearn/metrics/_regression.py:483: FutureWarning:

'squared' is deprecated in version 1.4 and will be removed in 1.6. To calculate the root mean squared error, use the function'root_mean_squared_error'.

# Замер времени обучения лучшей модели

start_time = time.time()
time_model_lgb = best_model_lgb
time_model_lgb.fit(X_train, y_train)
training_time_lgb = time.time() - start_time

# Замер времени предсказания лучшей модели

start_time = time.time()
y_pred_lgb = time_model_lgb.predict(X_train)
prediction_time_lgb = time.time() - start_time

print('LightGBM model training time, s:', training_time_lgb)
print('LightGBM model prediction time, s:', prediction_time_lgb)

LightGBM model training time, s: 1.1269159317016602
LightGBM model prediction time, s: 0.027966022491455078

# Визуализация y_true, y_pred, resid и их разницы для наглядности

fig = go.Figure()

fig.add_trace(go.Scatter(x=X_train.index, y=y_train,
                    mode='lines',
                    name='Original data'))

fig.add_trace(go.Scatter(x = X_train.index, y = y_pred_train_lgb,
                    mode='lines',
                    name='CB prediction'))

fig.add_trace(go.Scatter(x = X_train.index, y = decomposed.resid.loc[X_train.index],
                    mode='lines',
                    name='Resid'))

fig.add_trace(go.Scatter(x = X_train.index, y = decomposed.resid.loc[X_train.index] + y_pred_train_lgb.mean() - y_pred_train_lgb,
                    mode='lines',
                    name='Resid-error'))

fig.update_layout(xaxis_tickformat='%Y-%m-%d', xaxis_title='Datetime')
fig.update_layout(width=1000, height=800)
fig.show()

# Построение гистограммы остатков

plt.figure(figsize = (16,8))
plt.hist(y_train - y_pred_train_lgb, bins=100, label = 'model residual')
plt.hist(decomposed.resid.loc[X_train.index], bins=100, alpha = 0.7, label = 'residual component')
plt.xlabel('y_true - y_pred_train')
plt.ylabel('Frequency')
plt.axvline(x=0, color='r', linestyle='-', linewidth=1, label = 'zero')
plt.axvline(x=np.mean(y_train - y_pred_train_lgb), color='b', linestyle='-', linewidth=2, alpha = 0.2, label = 'residual mean')
plt.legend()
plt.title('LightGBM residual hist')
plt.grid('true')
plt.show();

# Поиск наиболее и наименее значимых фичей

feature_importances = best_model_lgb.feature_importances_

feature_importance_df = pd.DataFrame({
    'Feature': all_feature_names,
    'Importance': feature_importances
})

# Максимальные признаки

feature_importance_df_max = feature_importance_df.sort_values(
    by='Importance', ascending=False
)
plt.figure(figsize=(12, 8))
plt.barh(feature_importance_df_max['Feature'], feature_importance_df_max['Importance'], color='skyblue')
plt.xlabel('Importance')
plt.ylabel('Features')
plt.title('Maximal Feature Importance CatBoostRegressor')
plt.gca().invert_yaxis()
plt.show()

# Результаты обучения и получение RMSE метрик на тестовом и тренировочном наборах

y_pred_test_cb = best_model_cb.predict(X_test)
rmse_test_cb = mean_squared_error(y_test, y_pred_test_cb, squared = False)
print("Best CatBoostRegressor model Test RMSE:", rmse_test_cb)
print("Best CatBoostRegressor model Test mean residual:", np.mean(y_test - y_pred_test_cb))

Best CatBoostRegressor model Test RMSE: 44.039583298055476
Best CatBoostRegressor model Test mean residual: 14.394662367664692

/Users/NOrlov/myenv/lib/python3.12/site-packages/sklearn/metrics/_regression.py:483: FutureWarning:

'squared' is deprecated in version 1.4 and will be removed in 1.6. To calculate the root mean squared error, use the function'root_mean_squared_error'.

# Визуализация y_true, y_pred, resid и их разницы для наглядности

fig = go.Figure()

fig.add_trace(go.Scatter(x=X_test.index, y=y_test,
                    mode='lines',
                    name='Original data'))

fig.add_trace(go.Scatter(x = X_test.index, y = y_pred_test_cb,
                    mode='lines',
                    name='CB prediction'))

fig.add_trace(go.Scatter(x = X_test.index, y = decomposed.resid.loc[X_test.index],
                    mode='lines',
                    name='Resid'))

fig.add_trace(go.Scatter(x = X_test.index, y = decomposed.resid.loc[X_test.index] + y_pred_test_cb.mean() - y_pred_test_cb,
                    mode='lines',
                    name='Resid-error'))

fig.update_layout(xaxis_tickformat='%Y-%m-%d', xaxis_title='Datetime')
fig.update_layout(width=1000, height=800)
fig.show()

# Построение гистограммы остатков на Test наборе

plt.figure(figsize = (16,8))
plt.hist(y_test - y_pred_test_cb, bins=100, label = 'model residual')
plt.hist(decomposed.resid.loc[X_test.index], bins=100, alpha = 0.7, label = 'residual component')
plt.xlabel('y_true - y_pred_test')
plt.ylabel('Frequency')
plt.axvline(x=0, color='r', linestyle='-', linewidth=1, label = 'zero')
plt.axvline(x=np.mean(y_test - y_pred_test_cb), color='b', linestyle='-', linewidth=2, alpha = 0.2, label = 'residual mean')
plt.legend()
plt.title('CatBoostRegressor residual hist')
plt.grid('true')
plt.show();

	num_orders
datetime
2018-07-07 00:00:00	190
2018-03-17 20:00:00	57
2018-04-04 11:00:00	66
2018-06-05 18:00:00	60
2018-04-30 10:00:00	120

	num_orders
count	4416.000000
mean	84.422781
std	45.023853
min	0.000000
25%	54.000000
50%	78.000000
75%	107.000000
max	462.000000

	num_orders
datetime
2018-08-26 00:00:00	281
2018-08-18 02:00:00	254
2018-08-31 04:00:00	268
2018-08-18 00:00:00	281
2018-04-24 00:00:00	253

	num_orders	hour	day	dayofweek	lag_1	lag_2	lag_3	lag_4	lag_5	lag_6	lag_7	lag_8	rolling_mean
datetime
2018-06-17 20:00:00	71	20	17	6	78.0	37.0	124.0	156.0	91.0	91.0	50.0	88.0	80.291667
2018-05-13 07:00:00	13	7	13	6	23.0	24.0	61.0	155.0	71.0	59.0	95.0	71.0	65.041667
2018-06-10 04:00:00	88	4	10	6	106.0	101.0	83.0	194.0	114.0	110.0	88.0	119.0	90.125000
2018-07-03 19:00:00	95	19	3	1	78.0	108.0	130.0	95.0	79.0	58.0	81.0	93.0	85.958333
2018-07-10 18:00:00	74	18	10	1	108.0	120.0	80.0	97.0	77.0	72.0	98.0	103.0	90.416667
2018-07-17 17:00:00	120	17	17	1	57.0	97.0	74.0	49.0	66.0	121.0	79.0	108.0	86.833333
2018-04-02 11:00:00	82	11	2	0	111.0	57.0	62.0	7.0	20.0	33.0	33.0	47.0	62.791667
2018-03-09 04:00:00	30	4	9	4	31.0	29.0	66.0	84.0	79.0	88.0	86.0	114.0	63.125000
2018-03-11 23:00:00	118	23	11	6	89.0	119.0	99.0	99.0	107.0	78.0	93.0	60.0	64.791667
2018-04-30 14:00:00	49	14	30	0	62.0	49.0	63.0	120.0	91.0	69.0	21.0	32.0	77.000000
2018-08-13 05:00:00	83	5	13	0	143.0	168.0	437.0	273.0	160.0	185.0	118.0	135.0	130.500000
2018-05-13 22:00:00	88	22	13	6	101.0	94.0	70.0	48.0	117.0	128.0	54.0	71.0	68.916667
2018-06-11 08:00:00	58	8	11	0	17.0	46.0	68.0	101.0	116.0	113.0	124.0	152.0	83.125000
2018-07-29 21:00:00	84	21	29	6	114.0	89.0	89.0	131.0	178.0	119.0	135.0	116.0	106.333333
2018-05-28 02:00:00	186	2	28	0	78.0	157.0	113.0	111.0	77.0	78.0	84.0	37.0	72.541667

Прогнозирование заказов такси¶

Подготовка¶

Выводы¶

Анализ¶

Первичный анализ¶

Анализ трендов и сезонности¶

Выводы¶

Feature engineering¶

Выводы¶

Обучение¶

Создание тестовой и тренировочной выборок¶

CatBoostRegressor model¶

CatBoostRegressor Выводы¶

LightGBM model¶

LightGBM Выводы¶

Выводы¶

Анализ модели¶

Выводы¶

Выводы¶